Store and process the Crossref Database

This repository downloads Crossref metadata using the Crossref API. The items retrieved are stored in MongoDB to preserve their raw structure. This design allows for flexible downstream analyses.

MongoDB

MongoDB is run via Docker. It's available on the host machine at http://localhost:27017/.

docker run \
  --name=mongo-crossref \
  --publish=27017:27017 \
  --volume=`pwd`/mongo.db:/data/db \
  --rm \
  mongo:3.4.2

Execution

works

With mongo running, execute with the following commands:

# Download all works
# To start fresh, use `--cursor=*`
# If querying fails midway, you can extract the cursor of the
# last successful query from the tail of query-works.log.
# Then rerun download.py, passing the intermediate cursor
# to --cursor instead of *.
python download.py \
  --component=works \
  --batch-size=550 \
  --log=logs/query-works.log \
  --cursor=*

# Export mongodb works collection to JSON
mongoexport \
  --db=crossref \
  --collection=works \
  | xz > data/mongo-export/crossref-works.json.xz

See data/mongo-export for more information on crossref-works.json.xz. Note that creating this file from the Crossref API takes several weeks. Users are encouraged to use the cached version available on figshare.

1.works-to-dataframe.ipynb is a Jupyter notebook that extracts tabular datasets of works (TSVs), which are tracked using Git LFS:

doi.tsv.xz: a table where each row is a work, with columns for the DOI, type, and issued date.
doi-to-issn.tsv.xz: a table where each row is a work (DOI) to journal (ISSN) mapping.

types

With mongo running, execute with the following command:

python download.py \
  --component=types \
  --log=logs/query-types.log

Environment

This repository uses conda to manage its environment as specified in environment.yml. Install the environment with:

conda env create --file=environment.yml

Then use source activate crossref and source deactivate to activate or deactivate the environment. On windows, use activate crossref and deactivate instead.

Acknowledgements

This work is funded in part by the Gordon and Betty Moore Foundation's Data-Driven Discovery Initiative through Grant GBMF4552 to @cgreene.